DATA-DRIVEN CONTROL SYSTEM BASED ON LIGHTGBM ON QUADCOPTER
Unmanned Aerial Vehicles (UAVs), or drones, are unmanned aircraft that can fly and are programmed to perform specific tasks without human intervention. One of these UAVs is the quadcopter. Control systems on quadcopters are developing very rapidly over time. The quadcopter control system can be p...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/84269 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | Unmanned Aerial Vehicles (UAVs), or drones, are unmanned aircraft that can fly
and are programmed to perform specific tasks without human intervention. One of
these UAVs is the quadcopter. Control systems on quadcopters are developing very
rapidly over time. The quadcopter control system can be performed by using
machine learning. Light Gradient Boost Machine (LightGBM) is a gradient tree-
based learning algorithm that can optimize memory usage and training time with
techniques that selectively retain objects with large amounts of data during
training. In this research, a LightGBM algorithm design will be carried out to
replace the Linear Quadratic Regulator (LQR) control algorithm. This research
schemes to obtain input and output data from the closed-loop quadcopter system
using the LQR control algorithm, and then training will be carried out to get a
model of the LightGBM control algorithm to be used. With the expectation that the
LightGBM control algorithm can replace the LQR algorithm as a controller for the
quadcopter system that has been designed. The results of the simulations carried
out in this study show that by using hyperparameters (num leaves: 31, max depth:
7, learning rate: 0.09848640149856627 and num iterations: 872) for data without
disturbance the response of the LQR and LightGBM controller algorithms have
almost the same response results, with the rise time, settling time, and % overshoot
of the LightGBM algorithm faster than the LQR algorithm. For data with
disturbances, training is carried out using the following hyperparameters
(n_estimators = 100, learning rate: 0.3764792620878137 and num iterations:
880), with the result that the response generated by the LightGBM controller has a
longer rise time, and settling time than the response generated by the LQR
algorithm. |
|---|